[graphics] Day/night live integration — deploy --lighting day variants + time-of-day plate swap on the box

[100yenadmin]

The day/night dimension is proven at generation (generate_room.py --lighting day|night, DAYNIGHT_church.png) but never integrated live. Deploy day variants for the adopted rooms, wire the swap via the existing _location_plates/_active mechanism (or a _time_of_day sentinel mirroring the established pattern), capture both states in live combat. Post-ship batch item, now tracked.

[100yenadmin]

G5b wiring — day/night plate swap, live-combat evidence

Deploy: the 3 QC'd day plates landed on the box alongside their night counterparts, force-imported (ImportAsset ForceUpdate), confirmed loadable at 1024×1024:

• Assets/painterly/backdrops/tavern_day_v1.png (alongside tavern_layered_v1.png)
• Assets/painterly/backdrops/church_day_v1.png (alongside church_layered_v1.png)
• Assets/painterly/backdrops/bosshall_day_v1.png (alongside bosshall_layered_v1.png)

Mechanism (PR #1298, unmerged — awaiting review): #1298

Checked first whether an engine-side time-of-day signal is already surfaced to the renderer: the engine tracks campaign.time_of_day (servers/engine/travel.py PHASES) but build_combat_surface (viewer/server.py) does not currently expose it on /combat-surface. So there's no existing field to piggyback on yet — the minimal honest wiring is a box-side sentinel file, mirroring the exact pattern _active_combat.txt/_location_plates.json already use:

• New _time_of_day.txt sentinel (box-side config, same directory as the existing plate configs).
• paint_combat_v1.cs's existing plate-resolution block (right after the _location_plates.json/_active_combat.txt precedence chain) gets one additive check: if the sentinel says day and a naming-convention sibling <room>_day_v1.png exists (room = the resolved plate's filename prefix up to the first _, e.g. church_layered_v1.png → church), prefer the day plate.
• Default-off / additive: no sentinel file, sentinel ≠ day, or a missing day plate → falls through unchanged to today's night/current plate. Verified this is a pure 19-line addition, no existing lines touched.
• A future pass could have the engine (or deploy_room.sh) write the sentinel automatically from campaign.time_of_day instead of a manually-set file — flagged as follow-up, not required for this wiring pass.

Live-combat capture (same scene, both states) — Cathedral Nave, camp_gfxchurch01 (seeded via qa/seed_gfx_church.py, live combat: Aldric vs Goblin Warrior at cells (6,8)/(7,5)):

• Night (baseline, no sentinel): ~/worldos-session-notes/renders/church_night_live_v1.png — mean luma 35.6, 56.8% non-black pixels. Torch-lit nave, same actor poses/ring markers.
• Day (_time_of_day.txt=day): ~/worldos-session-notes/renders/church_day_live_v1.png — mean luma 117.3, 99.9% non-black pixels. Sunlit stained-glass nave, pews and altar clearly legible.
• Verified the mechanism directly (not just brightness correlation): a box-side readback after the day render returned sentinel=day backdropTex=church_day_v1 — the scene's actual loaded backdrop texture, confirming the plate-swap fired, not a coincidental re-render.
• Non-black gate: PASS both frames. Eyeball: day/night read as unmistakably distinct while the combat scene (both actors, same relative positions/selection rings) is identical — this is a genuine same-scene A/B, not two different rooms.

Box state: restored to the exact before-state after capture — _active_combat.txt→crypt_dense_v1.png, _active_campaign.txt→camp_gfxchurchcrypt01, _time_of_day.txt removed, no _location_plates.json (all match preflight). Editor pid 4241 unchanged (exactly one editor, legit /Hub/Editor/ path), :8080 bridge healthy (pid 2025 unchanged), :8765→Mac:8770 reverse tunnel healthy (200 on /combat-surface). Deployed day/night plates themselves are left in place per instructions.

Wiring half of #1291 is done: plate deploy + selection rule + same-scene day/night proof. PR #1298 is open for review before merge.